3.5 KiB
eip | title | description | author | discussions-to | status | type | category | created |
---|---|---|---|---|---|---|---|---|
3855 | PUSH0 instruction | Introduce a new instruction which pushes the constant value 0 onto the stack | Alex Beregszaszi (@axic), Hugo De la cruz (@hugo-dc), Paweł Bylica (@chfast) | https://ethereum-magicians.org/t/eip-3855-push0-instruction/7014 | Stagnant | Standards Track | Core | 2021-02-19 |
Abstract
Introduce the PUSH0
(0x5f
) instruction, which pushes the constant value 0 onto the stack.
Motivation
Many instructions expect offsets as inputs, which in a number of cases are zero. A good example is the return data parameters of CALLs
, which are set to zeroes in case the contract prefers using RETURNDATA*
. This is only one example, but there are many other reasons why a contract would need to push a zero value. They can achieve that today by PUSH1 0
, which costs 3 gas at runtime, and is encoded as two bytes which means 2 * 200
gas deployment cost.
Because of the overall cost many try to use various other instructions to achieve the same effect. Common examples include PC
, MSIZE
, CALLDATASIZE
, RETURNDATASIZE
, CODESIZE
, CALLVALUE
, and SELFBALANCE
. Some of these cost only 2 gas and are a single byte long, but their value can depend on the context.
We have conducted an analysis on Mainnet (block ranges 8,567,259…8,582,058 and 12,205,970…12,817,405), and ~11.5% of all the PUSH*
instructions executed push a value of zero.
The main motivations for this change include:
- Reducing contract code size.
- Reducing the risk of contracts (mis)using various instructions as an optimisation measure. Repricing/changing those instructions can be more risky.
- Reduce the need to use
DUP
instructions for duplicating zeroes.
To put the "waste" into perspective, across existing accounts 340,557,331 bytes are wasted on PUSH1 00
instructions, which means 68,111,466,200 gas was spent to deploy them. In practice a lot of these accounts share identical bytecode with others, so their total stored size in clients is lower, however the deploy time cost must have been paid nevertheless.
An example for 2) is changing the behaviour of RETURNDATASIZE
such that it may not be guaranteed to be zero at the beginning of the call frame. This was proposed as a way to chain transactions (i.e. EIP-2733).
Specification
The instruction PUSH0
is introduced at 0x5f
. It has no immediate data, pops no items from the stack, and places a single item with the value 0 onto the stack. The cost of this instruction is 2 gas (aka base
).
Rationale
Gas cost
The base
gas cost is used for instructions which place constant values onto the stack, such as ADDRESS
, ORIGIN
, and so forth.
Opcode
0x5f
means it is in a "contiguous" space with the rest of the PUSH
implementations and potentially could share the implementation.
Backwards Compatibility
This EIP introduces a new opcode which did not exists previously. Already deployed contracts using this opcode could change their behaviour after this EIP.
Test Cases
5F
-- successful execution, stack consist of a single item, set to zero5F5F..5F
(1024 times) -- successful execution, stack consists of 1024 items, all set to zero5F5F..5F
(1025 times) -- execution aborts due to out of stack
Security Considerations
The authors are not aware of any impact on security. Note that jumpdest-analysis is unaffected, as PUSH0
has no immediate data bytes.
Copyright
Copyright and related rights waived via CC0.